Five-year supply variance showed no clear excess over random annual order in two Great Lakes reconstructions, 1950-2022

We tested whether dependence across years raised the variance of five-year mean water supply in Lake Superior and Michigan-Huron. We used L2SWBM V3.0 monthly component medians for 1950-2022 and formed supply as precipitation plus runoff minus evaporation. Five times the variance of overlapping five-year means, divided by annual variance, was 1.247 and 1.107. In 19,999 joint random permutations of annual order, one-sided p-values were 0.11775 and 0.23845. Neither test met the two-lake threshold of 0.025. Trend removal, separate source fitting periods and October-September years also gave no test below that threshold. This result extends a prior within-year covariance benchmark. It does not show that annual supply is independent, or that drought risk is unchanged. Overlap and finite record length reduce the exact permutation mean ratio to 0.9429 rather than one. Source balance constraints, uncertain component estimates, two selected lakes and 73 years limit inference. Code, monthly inputs, source hashes, all primary permutations, an audit and a figure are included.

Great Lakes · net basin supply · five-year variance · annual dependence · permutation test · null result